[HW Accel Support]: Crashing ffmpeg processes when using AMD iGPU parallel in other Container #18732
Replies: 8 comments 6 replies
-
|
You may need to increase the memory allocated to the iGPU |
Beta Was this translation helpful? Give feedback.
-
|
Hi Nick, Still exactly the same issue. Many thanks! |
Beta Was this translation helpful? Give feedback.
-
|
I am also experiencing this problem. My environment:
It crashes so hard that I can't even successfully execute I have tried the following with no improvement:
One time I was watching VRAM usage as it crashed and I did not see it climbing or changing in an unusual way. The problem goes away if I disable hardware acceleration via: I recently migrated from an Intel N100 and I had no such problem on the N100. |
Beta Was this translation helpful? Give feedback.
-
|
I have done some more troubleshooting. Single vs multiple streamsIf I use ❌ VRAMI tried increasing VRAM to 16GB. No improvement. ❌ Host backportsNext I tried updating several debian packages on my host from trixie-backports: Resulting in: At first it seemed like that might of helped, but eventually it crashed again. ✅ Container backportsFinally I have tried updating mesa drivers inside the container: Contents of So far Frigate has been stable using this updated image. Next stepsIf it remains stable, I will roll back my other changes (like host backports and VRAM increases) to see if this single change is sufficient to keep Frigate stable. |
Beta Was this translation helpful? Give feedback.
-
|
Good tests and thanks for your investigation! I can also test it on my instance, once I'm back from holiday. BR |
Beta Was this translation helpful? Give feedback.
-
|
I think that I isolated the solution as being backported drivers inside the container. I have undone the following changes and Frigate is still stable:
|
Beta Was this translation helpful? Give feedback.
-
|
I would like to try your solution but I am not quite sure which changes I effectively need to make to the base image. Could you clarify the changes needed to the base image? |
Beta Was this translation helpful? Give feedback.
-
|
If you're using a compose file for your services, use something like: Compose FileAssuming you're using ./data/frigate as the base folder for frigate, you have to create 2 files: 1. Dockerfile2. backports.sourcesThen in your compose add this line to the frigate service: |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the tip @kRYstall9 , the solution to build a new image and update the drivers completely solved this "amdgpu_device_initialize: amdgpu_query_gpu_info_init failed" error message I had when passing through my Ryzen 7900x iGPU. As a bonus, I have no more ffmpeg frame errors either. My log are error free! |
Beta Was this translation helpful? Give feedback.
-
|
I'm interested in the possible fixes for this as I'm pretty sure I experienced this problem with my AMD 7800 processor. I'm running in an LXC in Proxmox, which I know is not technically supported, but the docker container died a horrible death locking up the LXC and causing issues on the host. Eventually ended up manually killing everything and rebooting the host. |
Beta Was this translation helpful? Give feedback.
-
|
Just tried the "container backport" hack but it made no difference. Still death lock after ~1 day. I'm on Ryzen 9 7940HS w/ Radeon 780M on Ubuntu 24.04 host, kernel 6.16. |
Beta Was this translation helpful? Give feedback.
-
|
I'm on GB-BER7-8840 - Ryzen 7 Pro 8840/AMD Radeon 780M - 48GB DDR and 16GB reserved for GPU - runing Ubuntu. I just modified the dockerfile: So far so good, thank you for tip @kRYstall9 |
Beta Was this translation helpful? Give feedback.
-
|
What's the error you're getting? I fixed all the errors by running an "alpha" release of the v17 of frigate
N.B. This is not one of the latest versions but it's actually working for me. You can find newer versions here |
Beta Was this translation helpful? Give feedback.
-
|
ffmpeg is simply crashing, there is an error: Unable to poll radeon GPU stats: amdgpu_device_initialize: amdgpu_query_gpu_info_init failed and after that ffmpeg dies. It happens in 1-2hours after start. I will test v17, for now i simply went with full CPU config without hwaccel (and coral for detections) - that is working fine and as I have 9 cams total that Ryzen works on 7% load. @kRYstall9 nope - tested 5 or 6 alpha versions. It works fine for 1-2hours and then i get: and ffmpeg crashes. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Describe the problem you are having
Hi,
first of all thanks for this great project.
I have a strange issue with my setup consiting of a Ryzen 8840U with a Radeon 780M iGPU:
I use the frigate:0.15.1-rocm image and preset-vaapi for hw acceleration, which is working fine unless I use the GPU in e.g. Immich, where it is configured in the machine learning container (and where it is working fine of course). Once the container is doing something the ffmpeg processes are crashing.
I already played with privileged true flag but this had no impact.
Could you help me here?
Many thanks!
Rainer
Version
0.15.0-6cb5cfb
Frigate config file
docker-compose file or Docker CLI command
Relevant Frigate log output
Relevant go2rtc log output
FFprobe output from your camera
Let me know if relevantInstall method
Docker Compose
Object Detector
Coral
Network connection
Wired
Camera make and model
hikvision DS-2CD2087G2-LU
Screenshots of the Frigate UI's System metrics pages
Any other information that may be helpful
No response
Beta Was this translation helpful? Give feedback.
All reactions